Skip to main content

​
Instance Types

TensorPool offers a variety of GPU instance types that can be used for both clusters and jobs.
Instance TypeMulti-Node
1xB200No
2xB200No
4xB200No
8xB200Yes
1xH200No
2xH200No
4xH200No
8xH200Yes
1xH100No
2xH100No
4xH100No
8xH100Yes
1xMI300XNo
2xMI300XNo
4xMI300XNo

​
Where to Use Instance Types

  • Clusters: Long-running infrastructure for development, training, and experimentation
  • Jobs: Single-run compute tasks with automatic cleanup after completion

​
Choosing the Right Instance Type

​
Single GPU (1x)

Best for development, debugging, small models, and inference. Most cost-effective for getting started.

​
Multi-GPU Single Node (2x, 4x, 8x)

For training larger models with data parallelism or model parallelism

​
Multi-Node Clusters

Best for large models, massive datasets, and distributed training. Currently supported for 8xB200 & 8xH200 only.

​
Examples

​
Creating Clusters

# Single H200 node
tp cluster create -t 1xB200

# Multi-node: 4 nodes with 8xB200 each (32 GPUs total)
tp cluster create -t 8xB200 -n 4

​
Running Jobs

# tp.config.toml - Single H100 job
commands = ["python train.py"]
instance_type = "1xH100"
outputs = ["checkpoints/"]
ignore = [".git", ".venv"]

tp job push tp.config.toml

​
Next Steps

⌘I